Implementing Natural Language Cognitive Architecture with GPT-3 and the "nexus" concept

Key Takeaways

hey everyone david shapiro here uh i just have a really quick update i wanted to share an insight that i had about how to implement raven or natural language cognitive architecture um basically what i what i realized was um because raven's gonna run as an instance probably in a container i can actually use a shared file system for a lot of the memory and so what i mean by that is well let me just show you so i unfortunately i can't have the services running because i'm using the audio services of my computer but basically what what it does is the microphone will capture a sequence of audio files which can then be used for inferencing things like speaker recognition um speech to text that sort of thing and it'll be cached here and then used uh and then it'll be cleaned up so it'll just be a buffer and then it'll be translated into normal text here so you see um whoops there it is so you see uh it's speech-to-text uh and that's that's that's basically kind of the gist of it there's still a lot of services set up uh here's me drinking a beer so this is basically what raven sees and that will be translated into text via inference image based inference and so that's what i'm working on now is setting up these services then there's another thing so that's the audio cache and the video cache so that's basically the sensory buffer for an agi next is the heartbeat so you know how you can be aware of your own body it's called proprioception this is basically raven's proprioception the every service that runs is going to create a heartbeat file here and every every time every time that one of those files cycles it will update the the time stamp so this way raven will be aware of the services that are running there's a whole bunch of other services that i need to work on i don't want to spoil anything but they're coming so microphone and camera that's the time stamp that they were running last here's the services you can see it's uh what just 64 lines of python some of it is comments so these are these are very small services they just use file they use files to share and then what i'm going to set up next is the is the speech to text service and then a video inference or image object detection inference the advantage of having it set up this way let me go back to the file folder sorry the advantage to having it set up this way is that any number an arbitrary number of audio processing services can use these files so let's say in the long run i'm going to have music recognition ambient sound recognition emotional tone recognition there's going to be all kinds of services using these audio files to generate inferences and you see the fact that every audio file has a time stamp attached to it that means that raven will know exactly when he heard what so that's pretty cool and same with the video again there's me drinking a beer as i'm getting set up so right now the uh the frame rate for this is one frame per second or actually it's every two seconds um that's because it's expensive to run this stuff um and it would be excessive to run it at any higher rate in the future raven will be updating the vision at you know like one frame per second two frames per second ten frames per second and eventually you know as the technology gets cheaper and faster raven will see as fast as you or i do that might be five or ten years down the road and then the memories so everything that enters raven's consciousness the agis consciousness will end up as a log file here eventually i'm going to transfer this into a private encrypted blockchain because they're they're chronologically sequential everything that raven thinks sees hears says everything is going to end up here in the memories directory and then it's going to be encrypted and put in a blockchain for privacy's sake but for now i'm just using log files i'm thinking about using something like syslog there's a there's a technology called the elk stack which is elasticsearch log stash and kibana which is a tool for capturing and visualizing logs because that's essentially what the content of raven's consciousness is is a series of log files it's that simple and i know all of this looks really simple it's not complete yet so keep that in mind and it is you might think of it as elegantly simple but i also think of it as deceptively simple because just because something is complex or next generation right that doesn't mean that it's that it's uh that it has to be complex or or over the top right simple solutions are are often the best solutions so anyways that's it i just wanted to post a quick update about this thanks for watching bye